August 2019
Intermediate to advanced
202 pages
5h 9m
English
One of the major limitations of the process described in the preceding section is that it doesn't include the training process. This can be a major impediment in use cases that involve checkpointing at some point during the training process. To overcome it, TensorFlow makes it possible to save models in their entirety. This can primarily be achieved in two ways—using the Keras API or using the SavedModel API.
In the following sections, we briefly discuss both methods and their syntax. We also provide insights into when to use each.
Read now
Unlock full access